When trying a new model in internal systems, it is natural to use the announced price for the first estimate. But if that price is time-limited, and the discount is not deducted from the bill but refunded later as credits, monthly bills and budgets look different.
On October 13, 2026, Google announced that it would offer a new model, Gemini 4 Argon, on Google Cloud. The announcement blog post lists an introductory price through November 30 of $2 per million input tokens and $10 for output. In this article, we check how that price is described on the Google Cloud (Agent Platform) and Gemini API pricing pages, and lay out the points to keep in mind before estimating costs or seeking internal approval.
Pricing and availability in the announcement blog post
The official Google Cloud blog post (October 13, 2026) describes availability and pricing as follows.
Starting today, Gemini 4 Argon is available in Gemini for business on the web, desktop, and mobile. Enterprise developers and builders can access Argon via API in Gemini platform under model ID gemini-4-argon, in Google AI Studio, and in Google Antigravity.
Argon launches at an introductory rate of $2 per million input tokens and $10 per million output tokens through November 30, 2026, with cached input tokens priced at 95% off. After November 30, standard pricing of $4 per million input tokens and $20 per million output tokens will apply.
The model ID for API use is gemini-4-argon. Through November 30, it costs $2 for input and $10 for output; after that, $4 for input and $20 for output, exactly double. Cached input is stated to be 95% off. The Gemini API release notes also state that it was released as generally available (GA) on October 13, 2026.
The pricing page says "50% refunded as credits"
When we opened Google Cloud's generative AI pricing page (Agent Platform Pricing) on October 24, 2026, the Gemini 4 Argon row was split into two: "through November 30, 2026" and "Starting December 1, 2026". The row through November 30 carries a note.
Promotional pricing provided through 50% credits back on net spend on select models within a given period. Promotional pricing does not apply to Gemini 4 Argon Cyber Permissive.
In other words, the Agent Platform pricing page describes the introductory price as a mechanism that refunds 50% of net spend as credits. This page alone does not tell you whether the unit price itself is halved or refunded as credits after billing, or how it appears on invoices and credit statements. If you manage budgets, you need to check the billing details of your contracted account before looking at how the billed amount changes.
Also, this introductory price does not apply to "Gemini 4 Argon Cyber Permissive", which is listed under a separate name on the pricing page. On the pricing page, it sits in the same row as Argon from December 1 onward ($4 for input and $20 for output).
| Item (Standard, global) | Through November 30 | From December 1 |
|---|---|---|
| Input (per 1M tokens) | $2 | $4 |
| Cached input | $0.1 | $0.2 |
| Output (including reasoning) | $10 | $20 |
Endpoints other than global are each priced about 10% higher (for example, $2.2 for input and $11 for output). Besides Standard, the pricing page has Priority and Batch tabs, and input and output through November 30 were $3.6 and $18 for Priority, and $1 and $5 for Batch.
The pricing page for use with Google AI Studio or the Gemini API (ai.google.dev) also lists the same unit prices in the form "through November 30, 2026" and "starting December 1, 2026" (Standard at $2 for input and $10 for output, Priority at $3.6 and $18, and Batch at $1 and $5 are also the same). That page has no note about refunds as credits, and states that the model is not available on the free tier. Whether the credit form applies only via Agent Platform cannot be determined from the two pages. Check which channel you are contracted through, and then confirm with that channel's billing.
Billing for long outputs changes from November 13
Apart from pricing, the Agent Platform pricing page has one more note.
Gemini 4 can generate completions up to 1 million tokens in length. When a completion exceeds 32K tokens in length, users will be charged for retrievable token usage even if the request ultimately fails due to client or server timeouts. This will take effect on November 13 2026.
Gemini 4 can generate up to 1 million tokens of output. From November 13, 2026, if output exceeds 32,000 tokens, the tokens retrieved are billed even if the request fails due to a client or server timeout.
This note is on the Agent Platform pricing page, and we could not find the same statement on the Gemini API pricing page or the model description page. The Gemini API guide recommends using background execution for outputs over 64,000 tokens and setting the HTTP client timeout to 15 minutes or more.
If you have built API processes that generate long reports or summarize large volumes of material, you may have designed retries on the assumption that failures are not charged. From November 13 onward, partial output may be billed every time a request times out. The top of the pricing page also says that non-200 responses are not billed for either input or output, so this note can be read as an exception. The surest way to know how each response is treated is to check your actual bill.
What to check before estimating costs and seeking internal approval
What follows is the editorial team's analysis. If you try Argon in internal systems, checking the following in order will reduce miscalculations when you cross November 30.
- Estimate with the unit prices from December 1 onward: The price through November 30 is time-limited. Build estimates for production operation at $4 for input and $20 for output, and treat the period through November 30 separately as a trial-period discount
- Share how credits will appear with accounting: The Agent Platform pricing page describes the introductory price as 50% refunded as credits. Check the invoice amount and how the credits are applied on your first bill
- Set timeouts and limits for long-output processing: On Agent Platform, from November 13, output over 32,000 tokens may be billed even if it fails. Review the output limit (max output tokens) and the number of times the same process is retried after a timeout
- Compare with existing models under the same conditions: The blog says the model is strong at long multi-step tasks, but whether its accuracy justifies its cost on your own documents is a separate question. Run it side by side with the Gemini 3 series and others on the same input, and compare including the number of output tokens
Why your bill does not automatically go down even when a new model's unit price falls is covered in A cheaper new model will not automatically lower your bill. For choosing between it and the Gemini 3.6 Flash series, see Three new Gemini models arrive, and for setting limits when using BigQuery AI functions, see Daily token limits for BigQuery AI functions.
Scope of verification
On October 24, 2026, we opened and checked the official Google Cloud blog post "Gemini 4 Argon available on Google Cloud, built to help solve your business’ hardest problems" (October 13, 2026) and the generative AI pricing page (Agent Platform Pricing) directly on cloud.google.com (document research). Before publication, on October 27, 2026, we reopened both to confirm that the prices and notes had not changed, and also checked the Gemini API pricing page, release notes, model description page, and the Gemini 4 Argon guide on ai.google.dev. We have not checked the admin help for Gemini for business. The editorial team has not called Argon and has not verified its accuracy, performance in Japanese, or how invoices and credits actually appear. The Priority and Batch figures on the Agent Platform pricing page were read from the order of the tabs, and we confirmed the same figures, labeled by name, on the Gemini API pricing page.
For choosing AI models or estimating costs for internal systems, contact us through Development, AI, and automation consultations.
References
- Gemini 4 Argon available on Google Cloud, built to help solve your business’ hardest problems（Google Cloud Blog, 2026-10-13）
- Agent Platform Pricing（Google Cloud）
- Gemini Developer API pricing（Google AI for Developers）
- Gemini API release notes（Google AI for Developers）
- What’s new in Gemini 4 Argon（Google AI for Developers）
